摘要
采用偏最小二乘法对95个煤炭样品的近红外光谱数据进行处理,并提取主成分.将提取的主成分与煤炭的发热量、灰分、挥发份、含硫量和全水分共同作为变量,进行系统聚类分析.将样品数据聚类为4组,同时剔除异常样本.对聚类后的各组数据采用多元散射校正、二阶导数、诺里斯导数平滑进行预处理,建立偏最小二乘定量分析模型.采用逐步筛选法,求得以发热量为变量的Bayes判别函数,交互验证结果表明判别函数稳定性良好.对未知样品发热量、灰分、挥发份、含硫量和全水分预测的决定系数分别达到0.992、0.927、0.938、0.778、0.978,说明模型预测性能良好.
Partial least square（PLS） method was used to treat the near-infrared spectra data of 95 coal samples and their principal components were extracted.Taking the principal components combined with the coal calorific capacity,ash content,volatility,sulphur content,and overall moisture copacity as variables,the samples were hierarchically clustered into four homogeneous clusters and,meantion,the abnormal samples were excluded.The spectrum in every cluster were preprocessed with second derivative and multiplicative scattering correction（MSC） and smoothed with Norris derivative filter,and thus the PLS quantitative analysis model was established.The succesive selection method was used to construct a Bayes discriminate function with the calorific capacily as its variable.The result of interactive validation indicated that the stabilization of this function was perfect.The prediction decision coefficient of above-mentioned five quantities of the coal samples were 0.992,0.927,0.938,0.778,0.978,showing that the predictive performance of the model was fine.
出处
《兰州理工大学学报》
CAS
北大核心
2012年第1期59-62,共4页
Journal of Lanzhou University of Technology
关键词
近红外光谱
煤炭
最小偏二乘
系统聚类
判别函数
near-infrared spectrum
coal
partial least square
hierarchical cluster
discriminate function
